net/netfilter/nfnetlink_queue_core.c

   1 /*
   2  * This is a module which is used for queueing packets and communicating with
   3  * userspace via nfnetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  * (C) 2007 by Patrick McHardy <kaber@trash.net>
   7  *
   8  * Based on the old ipv4-only ip_queue.c:
   9  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
  10  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17 #include <linux/module.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/init.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/slab.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/netfilter_ipv4.h>
  27 #include <linux/netfilter_ipv6.h>
  28 #include <linux/netfilter/nfnetlink.h>
  29 #include <linux/netfilter/nfnetlink_queue.h>
  30 #include <linux/list.h>
  31 #include <net/sock.h>
  32 #include <net/netfilter/nf_queue.h>
  33 #include <net/netfilter/nfnetlink_queue.h>
  34
  35 #include <linux/atomic.h>
  36
  37 #ifdef CONFIG_BRIDGE_NETFILTER
  38 #include "../bridge/br_private.h"
  39 #endif
  40
  41 #define NFQNL_QMAX_DEFAULT 1024
  42
  43 struct nfqnl_instance {
  44         struct hlist_node hlist;                /* global list of queues */
  45         struct rcu_head rcu;
  46
  47         int peer_pid;
  48         unsigned int queue_maxlen;
  49         unsigned int copy_range;
  50         unsigned int queue_dropped;
  51         unsigned int queue_user_dropped;
  52
  53
  54         u_int16_t queue_num;                    /* number of this queue */
  55         u_int8_t copy_mode;
  56         u_int32_t flags;                        /* Set using NFQA_CFG_FLAGS */
  57 /*
  58  * Following fields are dirtied for each queued packet,
  59  * keep them in same cache line if possible.
  60  */
  61         spinlock_t      lock;
  62         unsigned int    queue_total;
  63         unsigned int    id_sequence;            /* 'sequence' of pkt ids */
  64         struct list_head queue_list;            /* packets in queue */
  65 };
  66
  67 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  68
  69 static DEFINE_SPINLOCK(instances_lock);
  70
  71 #define INSTANCE_BUCKETS        16
  72 static struct hlist_head instance_table[INSTANCE_BUCKETS] __read_mostly;
  73
  74 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  75 {
  76         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
  77 }
  78
  79 static struct nfqnl_instance *
  80 instance_lookup(u_int16_t queue_num)
  81 {
  82         struct hlist_head *head;
  83         struct hlist_node *pos;
  84         struct nfqnl_instance *inst;
  85
  86         head = &instance_table[instance_hashfn(queue_num)];
  87         hlist_for_each_entry_rcu(inst, pos, head, hlist) {
  88                 if (inst->queue_num == queue_num)
  89                         return inst;
  90         }
  91         return NULL;
  92 }
  93
  94 static struct nfqnl_instance *
  95 instance_create(u_int16_t queue_num, int pid)
  96 {
  97         struct nfqnl_instance *inst;
  98         unsigned int h;
  99         int err;
 100
 101         spin_lock(&instances_lock);
 102         if (instance_lookup(queue_num)) {
 103                 err = -EEXIST;
 104                 goto out_unlock;
 105         }
 106
 107         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 108         if (!inst) {
 109                 err = -ENOMEM;
 110                 goto out_unlock;
 111         }
 112
 113         inst->queue_num = queue_num;
 114         inst->peer_pid = pid;
 115         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 116         inst->copy_range = 0xfffff;
 117         inst->copy_mode = NFQNL_COPY_NONE;
 118         spin_lock_init(&inst->lock);
 119         INIT_LIST_HEAD(&inst->queue_list);
 120
 121         if (!try_module_get(THIS_MODULE)) {
 122                 err = -EAGAIN;
 123                 goto out_free;
 124         }
 125
 126         h = instance_hashfn(queue_num);
 127         hlist_add_head_rcu(&inst->hlist, &instance_table[h]);
 128
 129         spin_unlock(&instances_lock);
 130
 131         return inst;
 132
 133 out_free:
 134         kfree(inst);
 135 out_unlock:
 136         spin_unlock(&instances_lock);
 137         return ERR_PTR(err);
 138 }
 139
 140 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 141                         unsigned long data);
 142
 143 static void
 144 instance_destroy_rcu(struct rcu_head *head)
 145 {
 146         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 147                                                    rcu);
 148
 149         nfqnl_flush(inst, NULL, 0);
 150         kfree(inst);
 151         module_put(THIS_MODULE);
 152 }
 153
 154 static void
 155 __instance_destroy(struct nfqnl_instance *inst)
 156 {
 157         hlist_del_rcu(&inst->hlist);
 158         call_rcu(&inst->rcu, instance_destroy_rcu);
 159 }
 160
 161 static void
 162 instance_destroy(struct nfqnl_instance *inst)
 163 {
 164         spin_lock(&instances_lock);
 165         __instance_destroy(inst);
 166         spin_unlock(&instances_lock);
 167 }
 168
 169 static inline void
 170 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 171 {
 172        list_add_tail(&entry->list, &queue->queue_list);
 173        queue->queue_total++;
 174 }
 175
 176 static void
 177 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 178 {
 179         list_del(&entry->list);
 180         queue->queue_total--;
 181 }
 182
 183 static struct nf_queue_entry *
 184 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 185 {
 186         struct nf_queue_entry *entry = NULL, *i;
 187
 188         spin_lock_bh(&queue->lock);
 189
 190         list_for_each_entry(i, &queue->queue_list, list) {
 191                 if (i->id == id) {
 192                         entry = i;
 193                         break;
 194                 }
 195         }
 196
 197         if (entry)
 198                 __dequeue_entry(queue, entry);
 199
 200         spin_unlock_bh(&queue->lock);
 201
 202         return entry;
 203 }
 204
 205 static void
 206 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 207 {
 208         struct nf_queue_entry *entry, *next;
 209
 210         spin_lock_bh(&queue->lock);
 211         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 212                 if (!cmpfn || cmpfn(entry, data)) {
 213                         list_del(&entry->list);
 214                         queue->queue_total--;
 215                         nf_reinject(entry, NF_DROP);
 216                 }
 217         }
 218         spin_unlock_bh(&queue->lock);
 219 }
 220
 221 static struct sk_buff *
 222 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 223                            struct nf_queue_entry *entry,
 224                            __be32 **packet_id_ptr)
 225 {
 226         sk_buff_data_t old_tail;
 227         size_t size;
 228         size_t data_len = 0;
 229         struct sk_buff *skb;
 230         struct nlattr *nla;
 231         struct nfqnl_msg_packet_hdr *pmsg;
 232         struct nlmsghdr *nlh;
 233         struct nfgenmsg *nfmsg;
 234         struct sk_buff *entskb = entry->skb;
 235         struct net_device *indev;
 236         struct net_device *outdev;
 237         struct nf_conn *ct = NULL;
 238         enum ip_conntrack_info uninitialized_var(ctinfo);
 239
 240         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 241                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 242                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 243                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 244 #ifdef CONFIG_BRIDGE_NETFILTER
 245                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 246                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 247 #endif
 248                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 249                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 250                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 251
 252         outdev = entry->outdev;
 253
 254         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 255         case NFQNL_COPY_META:
 256         case NFQNL_COPY_NONE:
 257                 break;
 258
 259         case NFQNL_COPY_PACKET:
 260                 if (entskb->ip_summed == CHECKSUM_PARTIAL &&
 261                     skb_checksum_help(entskb))
 262                         return NULL;
 263
 264                 data_len = ACCESS_ONCE(queue->copy_range);
 265                 if (data_len == 0 || data_len > entskb->len)
 266                         data_len = entskb->len;
 267
 268                 size += nla_total_size(data_len);
 269                 break;
 270         }
 271
 272         if (queue->flags & NFQA_CFG_F_CONNTRACK)
 273                 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 274
 275         skb = alloc_skb(size, GFP_ATOMIC);
 276         if (!skb)
 277                 return NULL;
 278
 279         old_tail = skb->tail;
 280         nlh = nlmsg_put(skb, 0, 0,
 281                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 282                         sizeof(struct nfgenmsg), 0);
 283         if (!nlh) {
 284                 kfree_skb(skb);
 285                 return NULL;
 286         }
 287         nfmsg = nlmsg_data(nlh);
 288         nfmsg->nfgen_family = entry->pf;
 289         nfmsg->version = NFNETLINK_V0;
 290         nfmsg->res_id = htons(queue->queue_num);
 291
 292         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 293         pmsg = nla_data(nla);
 294         pmsg->hw_protocol       = entskb->protocol;
 295         pmsg->hook              = entry->hook;
 296         *packet_id_ptr          = &pmsg->packet_id;
 297
 298         indev = entry->indev;
 299         if (indev) {
 300 #ifndef CONFIG_BRIDGE_NETFILTER
 301                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 302                         goto nla_put_failure;
 303 #else
 304                 if (entry->pf == PF_BRIDGE) {
 305                         /* Case 1: indev is physical input device, we need to
 306                          * look for bridge group (when called from
 307                          * netfilter_bridge) */
 308                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 309                                          htonl(indev->ifindex)) ||
 310                         /* this is the bridge group "brX" */
 311                         /* rcu_read_lock()ed by __nf_queue */
 312                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 313                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 314                                 goto nla_put_failure;
 315                 } else {
 316                         /* Case 2: indev is bridge group, we need to look for
 317                          * physical device (when called from ipv4) */
 318                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 319                                          htonl(indev->ifindex)))
 320                                 goto nla_put_failure;
 321                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 322                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 323                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 324                                 goto nla_put_failure;
 325                 }
 326 #endif
 327         }
 328
 329         if (outdev) {
 330 #ifndef CONFIG_BRIDGE_NETFILTER
 331                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 332                         goto nla_put_failure;
 333 #else
 334                 if (entry->pf == PF_BRIDGE) {
 335                         /* Case 1: outdev is physical output device, we need to
 336                          * look for bridge group (when called from
 337                          * netfilter_bridge) */
 338                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 339                                          htonl(outdev->ifindex)) ||
 340                         /* this is the bridge group "brX" */
 341                         /* rcu_read_lock()ed by __nf_queue */
 342                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 343                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 344                                 goto nla_put_failure;
 345                 } else {
 346                         /* Case 2: outdev is bridge group, we need to look for
 347                          * physical output device (when called from ipv4) */
 348                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 349                                          htonl(outdev->ifindex)))
 350                                 goto nla_put_failure;
 351                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 352                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 353                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 354                                 goto nla_put_failure;
 355                 }
 356 #endif
 357         }
 358
 359         if (entskb->mark &&
 360             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 361                 goto nla_put_failure;
 362
 363         if (indev && entskb->dev &&
 364             entskb->mac_header != entskb->network_header) {
 365                 struct nfqnl_msg_packet_hw phw;
 366                 int len = dev_parse_header(entskb, phw.hw_addr);
 367                 if (len) {
 368                         phw.hw_addrlen = htons(len);
 369                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 370                                 goto nla_put_failure;
 371                 }
 372         }
 373
 374         if (entskb->tstamp.tv64) {
 375                 struct nfqnl_msg_packet_timestamp ts;
 376                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 377                 ts.sec = cpu_to_be64(tv.tv_sec);
 378                 ts.usec = cpu_to_be64(tv.tv_usec);
 379
 380                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 381                         goto nla_put_failure;
 382         }
 383
 384         if (data_len) {
 385                 struct nlattr *nla;
 386                 int sz = nla_attr_size(data_len);
 387
 388                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
 389                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 390                         kfree_skb(skb);
 391                         return NULL;
 392                 }
 393
 394                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
 395                 nla->nla_type = NFQA_PAYLOAD;
 396                 nla->nla_len = sz;
 397
 398                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
 399                         BUG();
 400         }
 401
 402         if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
 403                 goto nla_put_failure;
 404
 405         nlh->nlmsg_len = skb->tail - old_tail;
 406         return skb;
 407
 408 nla_put_failure:
 409         if (skb)
 410                 kfree_skb(skb);
 411         net_err_ratelimited("nf_queue: error creating packet message\n");
 412         return NULL;
 413 }
 414
 415 static int
 416 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 417 {
 418         struct sk_buff *nskb;
 419         struct nfqnl_instance *queue;
 420         int err = -ENOBUFS;
 421         __be32 *packet_id_ptr;
 422         int failopen = 0;
 423
 424         /* rcu_read_lock()ed by nf_hook_slow() */
 425         queue = instance_lookup(queuenum);
 426         if (!queue) {
 427                 err = -ESRCH;
 428                 goto err_out;
 429         }
 430
 431         if (queue->copy_mode == NFQNL_COPY_NONE) {
 432                 err = -EINVAL;
 433                 goto err_out;
 434         }
 435
 436         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 437         if (nskb == NULL) {
 438                 err = -ENOMEM;
 439                 goto err_out;
 440         }
 441         spin_lock_bh(&queue->lock);
 442
 443         if (!queue->peer_pid) {
 444                 err = -EINVAL;
 445                 goto err_out_free_nskb;
 446         }
 447         if (queue->queue_total >= queue->queue_maxlen) {
 448                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 449                         failopen = 1;
 450                         err = 0;
 451                 } else {
 452                         queue->queue_dropped++;
 453                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 454                                              queue->queue_total);
 455                 }
 456                 goto err_out_free_nskb;
 457         }
 458         entry->id = ++queue->id_sequence;
 459         *packet_id_ptr = htonl(entry->id);
 460
 461         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 462         err = nfnetlink_unicast(nskb, &init_net, queue->peer_pid, MSG_DONTWAIT);
 463         if (err < 0) {
 464                 queue->queue_user_dropped++;
 465                 goto err_out_unlock;
 466         }
 467
 468         __enqueue_entry(queue, entry);
 469
 470         spin_unlock_bh(&queue->lock);
 471         return 0;
 472
 473 err_out_free_nskb:
 474         kfree_skb(nskb);
 475 err_out_unlock:
 476         spin_unlock_bh(&queue->lock);
 477         if (failopen)
 478                 nf_reinject(entry, NF_ACCEPT);
 479 err_out:
 480         return err;
 481 }
 482
 483 static int
 484 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 485 {
 486         struct sk_buff *nskb;
 487
 488         if (diff < 0) {
 489                 if (pskb_trim(e->skb, data_len))
 490                         return -ENOMEM;
 491         } else if (diff > 0) {
 492                 if (data_len > 0xFFFF)
 493                         return -EINVAL;
 494                 if (diff > skb_tailroom(e->skb)) {
 495                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 496                                                diff, GFP_ATOMIC);
 497                         if (!nskb) {
 498                                 printk(KERN_WARNING "nf_queue: OOM "
 499                                       "in mangle, dropping packet\n");
 500                                 return -ENOMEM;
 501                         }
 502                         kfree_skb(e->skb);
 503                         e->skb = nskb;
 504                 }
 505                 skb_put(e->skb, diff);
 506         }
 507         if (!skb_make_writable(e->skb, data_len))
 508                 return -ENOMEM;
 509         skb_copy_to_linear_data(e->skb, data, data_len);
 510         e->skb->ip_summed = CHECKSUM_NONE;
 511         return 0;
 512 }
 513
 514 static int
 515 nfqnl_set_mode(struct nfqnl_instance *queue,
 516                unsigned char mode, unsigned int range)
 517 {
 518         int status = 0;
 519
 520         spin_lock_bh(&queue->lock);
 521         switch (mode) {
 522         case NFQNL_COPY_NONE:
 523         case NFQNL_COPY_META:
 524                 queue->copy_mode = mode;
 525                 queue->copy_range = 0;
 526                 break;
 527
 528         case NFQNL_COPY_PACKET:
 529                 queue->copy_mode = mode;
 530                 /* we're using struct nlattr which has 16bit nla_len */
 531                 if (range > 0xffff)
 532                         queue->copy_range = 0xffff;
 533                 else
 534                         queue->copy_range = range;
 535                 break;
 536
 537         default:
 538                 status = -EINVAL;
 539
 540         }
 541         spin_unlock_bh(&queue->lock);
 542
 543         return status;
 544 }
 545
 546 static int
 547 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 548 {
 549         if (entry->indev)
 550                 if (entry->indev->ifindex == ifindex)
 551                         return 1;
 552         if (entry->outdev)
 553                 if (entry->outdev->ifindex == ifindex)
 554                         return 1;
 555 #ifdef CONFIG_BRIDGE_NETFILTER
 556         if (entry->skb->nf_bridge) {
 557                 if (entry->skb->nf_bridge->physindev &&
 558                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 559                         return 1;
 560                 if (entry->skb->nf_bridge->physoutdev &&
 561                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 562                         return 1;
 563         }
 564 #endif
 565         return 0;
 566 }
 567
 568 /* drop all packets with either indev or outdev == ifindex from all queue
 569  * instances */
 570 static void
 571 nfqnl_dev_drop(int ifindex)
 572 {
 573         int i;
 574
 575         rcu_read_lock();
 576
 577         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 578                 struct hlist_node *tmp;
 579                 struct nfqnl_instance *inst;
 580                 struct hlist_head *head = &instance_table[i];
 581
 582                 hlist_for_each_entry_rcu(inst, tmp, head, hlist)
 583                         nfqnl_flush(inst, dev_cmp, ifindex);
 584         }
 585
 586         rcu_read_unlock();
 587 }
 588
 589 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 590
 591 static int
 592 nfqnl_rcv_dev_event(struct notifier_block *this,
 593                     unsigned long event, void *ptr)
 594 {
 595         struct net_device *dev = ptr;
 596
 597         if (!net_eq(dev_net(dev), &init_net))
 598                 return NOTIFY_DONE;
 599
 600         /* Drop any packets associated with the downed device */
 601         if (event == NETDEV_DOWN)
 602                 nfqnl_dev_drop(dev->ifindex);
 603         return NOTIFY_DONE;
 604 }
 605
 606 static struct notifier_block nfqnl_dev_notifier = {
 607         .notifier_call  = nfqnl_rcv_dev_event,
 608 };
 609
 610 static int
 611 nfqnl_rcv_nl_event(struct notifier_block *this,
 612                    unsigned long event, void *ptr)
 613 {
 614         struct netlink_notify *n = ptr;
 615
 616         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 617                 int i;
 618
 619                 /* destroy all instances for this pid */
 620                 spin_lock(&instances_lock);
 621                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 622                         struct hlist_node *tmp, *t2;
 623                         struct nfqnl_instance *inst;
 624                         struct hlist_head *head = &instance_table[i];
 625
 626                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 627                                 if ((n->net == &init_net) &&
 628                                     (n->pid == inst->peer_pid))
 629                                         __instance_destroy(inst);
 630                         }
 631                 }
 632                 spin_unlock(&instances_lock);
 633         }
 634         return NOTIFY_DONE;
 635 }
 636
 637 static struct notifier_block nfqnl_rtnl_notifier = {
 638         .notifier_call  = nfqnl_rcv_nl_event,
 639 };
 640
 641 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 642         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 643         [NFQA_MARK]             = { .type = NLA_U32 },
 644         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 645         [NFQA_CT]               = { .type = NLA_UNSPEC },
 646 };
 647
 648 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 649         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 650         [NFQA_MARK]             = { .type = NLA_U32 },
 651 };
 652
 653 static struct nfqnl_instance *verdict_instance_lookup(u16 queue_num, int nlpid)
 654 {
 655         struct nfqnl_instance *queue;
 656
 657         queue = instance_lookup(queue_num);
 658         if (!queue)
 659                 return ERR_PTR(-ENODEV);
 660
 661         if (queue->peer_pid != nlpid)
 662                 return ERR_PTR(-EPERM);
 663
 664         return queue;
 665 }
 666
 667 static struct nfqnl_msg_verdict_hdr*
 668 verdicthdr_get(const struct nlattr * const nfqa[])
 669 {
 670         struct nfqnl_msg_verdict_hdr *vhdr;
 671         unsigned int verdict;
 672
 673         if (!nfqa[NFQA_VERDICT_HDR])
 674                 return NULL;
 675
 676         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 677         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 678         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 679                 return NULL;
 680         return vhdr;
 681 }
 682
 683 static int nfq_id_after(unsigned int id, unsigned int max)
 684 {
 685         return (int)(id - max) > 0;
 686 }
 687
 688 static int
 689 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 690                    const struct nlmsghdr *nlh,
 691                    const struct nlattr * const nfqa[])
 692 {
 693         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 694         struct nf_queue_entry *entry, *tmp;
 695         unsigned int verdict, maxid;
 696         struct nfqnl_msg_verdict_hdr *vhdr;
 697         struct nfqnl_instance *queue;
 698         LIST_HEAD(batch_list);
 699         u16 queue_num = ntohs(nfmsg->res_id);
 700
 701         queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).pid);
 702         if (IS_ERR(queue))
 703                 return PTR_ERR(queue);
 704
 705         vhdr = verdicthdr_get(nfqa);
 706         if (!vhdr)
 707                 return -EINVAL;
 708
 709         verdict = ntohl(vhdr->verdict);
 710         maxid = ntohl(vhdr->id);
 711
 712         spin_lock_bh(&queue->lock);
 713
 714         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 715                 if (nfq_id_after(entry->id, maxid))
 716                         break;
 717                 __dequeue_entry(queue, entry);
 718                 list_add_tail(&entry->list, &batch_list);
 719         }
 720
 721         spin_unlock_bh(&queue->lock);
 722
 723         if (list_empty(&batch_list))
 724                 return -ENOENT;
 725
 726         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 727                 if (nfqa[NFQA_MARK])
 728                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 729                 nf_reinject(entry, verdict);
 730         }
 731         return 0;
 732 }
 733
 734 static int
 735 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 736                    const struct nlmsghdr *nlh,
 737                    const struct nlattr * const nfqa[])
 738 {
 739         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 740         u_int16_t queue_num = ntohs(nfmsg->res_id);
 741
 742         struct nfqnl_msg_verdict_hdr *vhdr;
 743         struct nfqnl_instance *queue;
 744         unsigned int verdict;
 745         struct nf_queue_entry *entry;
 746         enum ip_conntrack_info uninitialized_var(ctinfo);
 747         struct nf_conn *ct = NULL;
 748
 749         queue = instance_lookup(queue_num);
 750         if (!queue)
 751
 752         queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).pid);
 753         if (IS_ERR(queue))
 754                 return PTR_ERR(queue);
 755
 756         vhdr = verdicthdr_get(nfqa);
 757         if (!vhdr)
 758                 return -EINVAL;
 759
 760         verdict = ntohl(vhdr->verdict);
 761
 762         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 763         if (entry == NULL)
 764                 return -ENOENT;
 765
 766         rcu_read_lock();
 767         if (nfqa[NFQA_CT] && (queue->flags & NFQA_CFG_F_CONNTRACK))
 768                 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
 769
 770         if (nfqa[NFQA_PAYLOAD]) {
 771                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
 772                 int diff = payload_len - entry->skb->len;
 773
 774                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 775                                  payload_len, entry, diff) < 0)
 776                         verdict = NF_DROP;
 777
 778                 if (ct)
 779                         nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
 780         }
 781         rcu_read_unlock();
 782
 783         if (nfqa[NFQA_MARK])
 784                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 785
 786         nf_reinject(entry, verdict);
 787         return 0;
 788 }
 789
 790 static int
 791 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 792                   const struct nlmsghdr *nlh,
 793                   const struct nlattr * const nfqa[])
 794 {
 795         return -ENOTSUPP;
 796 }
 797
 798 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 799         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 800         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 801 };
 802
 803 static const struct nf_queue_handler nfqh = {
 804         .name   = "nf_queue",
 805         .outfn  = &nfqnl_enqueue_packet,
 806 };
 807
 808 static int
 809 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 810                   const struct nlmsghdr *nlh,
 811                   const struct nlattr * const nfqa[])
 812 {
 813         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 814         u_int16_t queue_num = ntohs(nfmsg->res_id);
 815         struct nfqnl_instance *queue;
 816         struct nfqnl_msg_config_cmd *cmd = NULL;
 817         int ret = 0;
 818
 819         if (nfqa[NFQA_CFG_CMD]) {
 820                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 821
 822                 /* Commands without queue context - might sleep */
 823                 switch (cmd->command) {
 824                 case NFQNL_CFG_CMD_PF_BIND:
 825                         return nf_register_queue_handler(ntohs(cmd->pf),
 826                                                          &nfqh);
 827                 case NFQNL_CFG_CMD_PF_UNBIND:
 828                         return nf_unregister_queue_handler(ntohs(cmd->pf),
 829                                                            &nfqh);
 830                 }
 831         }
 832
 833         rcu_read_lock();
 834         queue = instance_lookup(queue_num);
 835         if (queue && queue->peer_pid != NETLINK_CB(skb).pid) {
 836                 ret = -EPERM;
 837                 goto err_out_unlock;
 838         }
 839
 840         if (cmd != NULL) {
 841                 switch (cmd->command) {
 842                 case NFQNL_CFG_CMD_BIND:
 843                         if (queue) {
 844                                 ret = -EBUSY;
 845                                 goto err_out_unlock;
 846                         }
 847                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 848                         if (IS_ERR(queue)) {
 849                                 ret = PTR_ERR(queue);
 850                                 goto err_out_unlock;
 851                         }
 852                         break;
 853                 case NFQNL_CFG_CMD_UNBIND:
 854                         if (!queue) {
 855                                 ret = -ENODEV;
 856                                 goto err_out_unlock;
 857                         }
 858                         instance_destroy(queue);
 859                         break;
 860                 case NFQNL_CFG_CMD_PF_BIND:
 861                 case NFQNL_CFG_CMD_PF_UNBIND:
 862                         break;
 863                 default:
 864                         ret = -ENOTSUPP;
 865                         break;
 866                 }
 867         }
 868
 869         if (nfqa[NFQA_CFG_PARAMS]) {
 870                 struct nfqnl_msg_config_params *params;
 871
 872                 if (!queue) {
 873                         ret = -ENODEV;
 874                         goto err_out_unlock;
 875                 }
 876                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 877                 nfqnl_set_mode(queue, params->copy_mode,
 878                                 ntohl(params->copy_range));
 879         }
 880
 881         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 882                 __be32 *queue_maxlen;
 883
 884                 if (!queue) {
 885                         ret = -ENODEV;
 886                         goto err_out_unlock;
 887                 }
 888                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 889                 spin_lock_bh(&queue->lock);
 890                 queue->queue_maxlen = ntohl(*queue_maxlen);
 891                 spin_unlock_bh(&queue->lock);
 892         }
 893
 894         if (nfqa[NFQA_CFG_FLAGS]) {
 895                 __u32 flags, mask;
 896
 897                 if (!queue) {
 898                         ret = -ENODEV;
 899                         goto err_out_unlock;
 900                 }
 901
 902                 if (!nfqa[NFQA_CFG_MASK]) {
 903                         /* A mask is needed to specify which flags are being
 904                          * changed.
 905                          */
 906                         ret = -EINVAL;
 907                         goto err_out_unlock;
 908                 }
 909
 910                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
 911                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
 912
 913                 if (flags >= NFQA_CFG_F_MAX) {
 914                         ret = -EOPNOTSUPP;
 915                         goto err_out_unlock;
 916                 }
 917
 918                 spin_lock_bh(&queue->lock);
 919                 queue->flags &= ~mask;
 920                 queue->flags |= flags & mask;
 921                 spin_unlock_bh(&queue->lock);
 922         }
 923
 924 err_out_unlock:
 925         rcu_read_unlock();
 926         return ret;
 927 }
 928
 929 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 930         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
 931                                     .attr_count = NFQA_MAX, },
 932         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
 933                                     .attr_count = NFQA_MAX,
 934                                     .policy = nfqa_verdict_policy },
 935         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 936                                     .attr_count = NFQA_CFG_MAX,
 937                                     .policy = nfqa_cfg_policy },
 938         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
 939                                     .attr_count = NFQA_MAX,
 940                                     .policy = nfqa_verdict_batch_policy },
 941 };
 942
 943 static const struct nfnetlink_subsystem nfqnl_subsys = {
 944         .name           = "nf_queue",
 945         .subsys_id      = NFNL_SUBSYS_QUEUE,
 946         .cb_count       = NFQNL_MSG_MAX,
 947         .cb             = nfqnl_cb,
 948 };
 949
 950 #ifdef CONFIG_PROC_FS
 951 struct iter_state {
 952         unsigned int bucket;
 953 };
 954
 955 static struct hlist_node *get_first(struct seq_file *seq)
 956 {
 957         struct iter_state *st = seq->private;
 958
 959         if (!st)
 960                 return NULL;
 961
 962         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 963                 if (!hlist_empty(&instance_table[st->bucket]))
 964                         return instance_table[st->bucket].first;
 965         }
 966         return NULL;
 967 }
 968
 969 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 970 {
 971         struct iter_state *st = seq->private;
 972
 973         h = h->next;
 974         while (!h) {
 975                 if (++st->bucket >= INSTANCE_BUCKETS)
 976                         return NULL;
 977
 978                 h = instance_table[st->bucket].first;
 979         }
 980         return h;
 981 }
 982
 983 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 984 {
 985         struct hlist_node *head;
 986         head = get_first(seq);
 987
 988         if (head)
 989                 while (pos && (head = get_next(seq, head)))
 990                         pos--;
 991         return pos ? NULL : head;
 992 }
 993
 994 static void *seq_start(struct seq_file *seq, loff_t *pos)
 995         __acquires(instances_lock)
 996 {
 997         spin_lock(&instances_lock);
 998         return get_idx(seq, *pos);
 999 }
1000
1001 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1002 {
1003         (*pos)++;
1004         return get_next(s, v);
1005 }
1006
1007 static void seq_stop(struct seq_file *s, void *v)
1008         __releases(instances_lock)
1009 {
1010         spin_unlock(&instances_lock);
1011 }
1012
1013 static int seq_show(struct seq_file *s, void *v)
1014 {
1015         const struct nfqnl_instance *inst = v;
1016
1017         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1018                           inst->queue_num,
1019                           inst->peer_pid, inst->queue_total,
1020                           inst->copy_mode, inst->copy_range,
1021                           inst->queue_dropped, inst->queue_user_dropped,
1022                           inst->id_sequence, 1);
1023 }
1024
1025 static const struct seq_operations nfqnl_seq_ops = {
1026         .start  = seq_start,
1027         .next   = seq_next,
1028         .stop   = seq_stop,
1029         .show   = seq_show,
1030 };
1031
1032 static int nfqnl_open(struct inode *inode, struct file *file)
1033 {
1034         return seq_open_private(file, &nfqnl_seq_ops,
1035                         sizeof(struct iter_state));
1036 }
1037
1038 static const struct file_operations nfqnl_file_ops = {
1039         .owner   = THIS_MODULE,
1040         .open    = nfqnl_open,
1041         .read    = seq_read,
1042         .llseek  = seq_lseek,
1043         .release = seq_release_private,
1044 };
1045
1046 #endif /* PROC_FS */
1047
1048 static int __init nfnetlink_queue_init(void)
1049 {
1050         int i, status = -ENOMEM;
1051
1052         for (i = 0; i < INSTANCE_BUCKETS; i++)
1053                 INIT_HLIST_HEAD(&instance_table[i]);
1054
1055         netlink_register_notifier(&nfqnl_rtnl_notifier);
1056         status = nfnetlink_subsys_register(&nfqnl_subsys);
1057         if (status < 0) {
1058                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1059                 goto cleanup_netlink_notifier;
1060         }
1061
1062 #ifdef CONFIG_PROC_FS
1063         if (!proc_create("nfnetlink_queue", 0440,
1064                          proc_net_netfilter, &nfqnl_file_ops))
1065                 goto cleanup_subsys;
1066 #endif
1067
1068         register_netdevice_notifier(&nfqnl_dev_notifier);
1069         return status;
1070
1071 #ifdef CONFIG_PROC_FS
1072 cleanup_subsys:
1073         nfnetlink_subsys_unregister(&nfqnl_subsys);
1074 #endif
1075 cleanup_netlink_notifier:
1076         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1077         return status;
1078 }
1079
1080 static void __exit nfnetlink_queue_fini(void)
1081 {
1082         nf_unregister_queue_handlers(&nfqh);
1083         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1084 #ifdef CONFIG_PROC_FS
1085         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1086 #endif
1087         nfnetlink_subsys_unregister(&nfqnl_subsys);
1088         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1089
1090         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1091 }
1092
1093 MODULE_DESCRIPTION("netfilter packet queue handler");
1094 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1095 MODULE_LICENSE("GPL");
1096 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1097
1098 module_init(nfnetlink_queue_init);
1099 module_exit(nfnetlink_queue_fini);